Learning to Self-Train for Semi-Supervised Few-Shot Classification

By Qianru Sun, Xinzhe Li, Yaoyao Liu, Shibao Zheng, Tat-seng Chua, Bernt Schiele. 2019

301 citations recorded by the source corpus.

Few-shot classification (FSC) is challenging due to the scarcity of labeled training data (e.g. only one labeled data point per class). Meta-learning has shown to achieve promising results by learning to initialize a classification model for FSC. In this paper we propose a novel semi-supervised meta-learning method called learning to self-train (LST) that leverages unlabeled data and specifically meta-learns how to cherry-pick and label such unsupervised data to further improve performance. To this end, we train the LST model through a large number of semi-supervised few-shot tasks. On each task, we train a few-shot model to predict pseudo labels for unlabeled data, and then iterate the self-training steps on labeled and pseudo-labeled data with each step followed by fine-tuning. We additionally learn a soft weighting network (SWN) to optimize the self-training weights of pseudo labels so that better ones can contribute more to gradient descent optimization. We evaluate our LST method on two ImageNet benchmarks for semi-supervised few-shot classification and achieve large improvements over the state-of-the-art method. Code is at this https URL.

Subfields: Artificial Intelligence, Radiology, Nuclear Medicine and Imaging, Meta & Continual Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Radiology, Nuclear Medicine and Imaging

In Meta & Continual Learning

Citations in the corpus

Towards Label-Efficient Incremental Learning: A Survey (2023, background): “However, a combination of different forms of supervision is common in non-incremental learning, such as the combination of self-and few-shot-supervision [Su et al. , 2020], or semi and few-shot-supervision [Ren et al. , 2018; Li et al. , 2019].”
Meta-Learning with Less Forgetting on Large-Scale Non-Stationary Task Distributions (2022, positive): “It has been shown that existing works [47,35,36] are particularly sensitive to OOD data.”
Adaptive Aggregation Networks for Class-Incremental Learning (2020, background): “Technically, the network parameters are updated at one level and the key hyperparameters are updated at another level [13,21,24,45,46,52].”
Mnemonics Training: Multi-Class Incremental Learning Without Forgetting (2020, background): “Meta-learning [5, 31, 35, 15] is another BOP in which a meta-learner is optimized subject to the optimality of the base-learner.”
Meta-Learning in Neural Networks: A Survey (2020, neutral): “This can be used in semi-supervised learning, for example to directly learn the unlabeled set’s labels to optimize validation set performance [146], or to train a label generation function [147].”
Learning Adaptive Classifiers Synthesis for Generalized Few-Shot Learning (2019, methodology): “Inspired by [26, 44], we consider a lightweight fine-tune step based on the synthesized classifier”
Meta-Transfer Learning Through Hard Tasks (2019, methodology): “Its task-agnostic property makes it possible to generalize to few-shot supervised/semisupervised learning as well as unsupervised reinforcement learning [5], [17], [18], [21], [22], [25]–[27].”
An Ensemble of Epoch-Wise Empirical Bayes for Few-Shot Learning (2019, neutral): “(3) Gradient descent based methods [1, 13,14,20,25,37,38,43,57,70,86] usually employ a meta-learner that learns to fast adapt an NN base-learner to a new task within a few optimization steps.”
An Overview of Deep Semi-Supervised Learning (2020, methodology): “In addition to image classiﬁcation, self-training was also successfully applied to a variety of tasks, such as semantic segmentation [7], text classiﬁcation [98, 79], machine translation [139, 64, 25, 65] and when learning from noisy data [154].”
Few-shot Image Generation with Elastic Weight Consolidation (2020, background): “This also leads to some semi-supervised learning methods [19].”
Adaptive Risk Minimization: A Meta-Learning Approach for Tackling Group Shift (2020, methodology): “Some other meta-learning methods adapt using both labeled and unlabeled data, either in the semi supervised learning setting [51, 70, 35] or the transductive learning setting [40, 1, 24].”
Extract Free Dense Labels from CLIP (2021, methodology): “Self-training has firstly become popular in the semisupervised classification task [33,29,35,43] and is also recently applied in the semi-supervised/zero-shot semantic segmentation settings [27,37,4,28,40,36,56,8].”
DeepEMD: Few-Shot Image Classification With Differentiable Earth Mover’s Distance and Structured Classifiers (2020, unknown): “Optimization-based methods [2, 5, 9–11, 18, 26, 31–33, 36, 38, 40–42, 44, 46, 47, 54, 55, 58, 64, 79] target at effectively adapting model parameters to new tasks in the low-shot regime.”
Part-aware Prototype Network for Few-shot Semantic Segmentation (2020, background): “There have been several recent attempts aiming to improve the few-shot learning by incorporating a set of unlabeled data, referred to as semi-supervised few-shot learning [24,16,1].”
A Comprehensive Survey of Few-shot Learning: Evolution, Applications, Challenges, and Opportunities (2022, background): “Δ -encoder[76] NIPS’18 Data Expansion mini-ImageNet* ✓ CP-AAN [66] NIPS’18 Data Expansion Imagenet ✓ LST [65] NIPS’19 Data Expansion mini-ImageNet* ✓”
Few-Shot Learning With a Strong Teacher (2021, background): “, the learning rate [61], [13]; an image generator could also be meta-learned to augment the support set [62] and auxiliary self-supervised tasks are also incorporated during meta-training to improve the generalization ability of the model [63], [64], [65].”
Contextualizing Multiple Tasks via Learning to Decompose (2021, background): “Initialization of deep neural network [30, 15], instance embeddings [41, 29, 45, 38, 19], optimization policies [35, 19] and data augmentation [56, 22] could be learned across sampled tasks.”